Independent Banker - November 2024 - 46

6 AI-specific risks to watch
Some AI risks, such as an application that hallucinates or falls in love with the
human being using it, have captured the popular imagination. Risks like these
are obviously attention-getting but aren't necessarily relevant to bankers' work.
Here are six bank-centered risks worth considering before dabbling in AI.
AI risk #1: An inability to explain decision-making
" AI can't just be a magic box out there, making
decisions for your bank, " says Beth Sumner, vice
president of customer success at Finosec, an
automated cybersecurity solution. She urges
community bankers to cultivate a clear, high-level
understanding of which criteria and data an AI
platform is using.
Sumner has observed bank off icers who are
intimidated by AI, convinced they can't understand
how it works. Whatever the source of a laissez-faire
attitude, the risk is that a lack of " explainability " will
tarnish a bank's reputation with customers who
question a decision, as well as with regulators who
are tasked with ensuring lending decisions are made
fairly, consistently and without bias.
2
3
AI risk #3: Bias (or the perception of bias)
AI programs that assess loan applications can yield amazing results, but some
decisions have been called into question over concerns about bias.
Jason Henrichs, CEO of banking consortium Alloy Labs Alliance, notes
that today's AI models can be asked fairly open-ended questions, such as
what factors make for good credit underwriting in auto loans, and then detect
patterns and write their own algorithms. Problems occur when an algorithm
identifi es individuals with certain ethnicities or ZIP codes as riskier prospects
and charges these individuals higher rates.
Ironically, AI may hold the key to solving the problem of its own propensity
for biased decision-making. Henrichs explains that an AI system can be taught
what " bias " means and then asked whether its outcomes are tainted by bias. In
other words, he says, AI may be an excellent source of information on whether
bias has infl uenced how AI-generated choices are made.
46 // ICBA Independent Banker // November 2024
AI risk #2: Ignorance of
third-party exposures
For most community banks, coming
to terms with AI means coming to
terms with how third-party vendors
are using AI, says Finosec's Sumner.
" Realistically, very few bankers
are going to develop or design their
own AI systems, " she says. " Instead,
this is going to be a third-party, riskmanagement
issue. "
One thing that makes managing
how third parties use AI so
challenging for community banks
is that vendors' practices are in a
state of constant flux. It's therefore
worth repeatedly asking third
parties if and when AI will be
incorporated into their product
suite, and whether models are
being trained on your bank's
proprietary information.
Overseeing third-party
relationships is an area where some
community banks are beginning
to shine. At the beginning of 2024,
David Robinson of Academy Bank
proactively approached his bank's
critical vendors to record and
better understand their use of AI.
He also harnessed AI to scan news
articles about the bank's existing
or potential third parties, flagging
references to their AI use and to
any negative incidents reported by
the media.
Independent Banker - November 2024

Table of Contents for the Digital Edition of Independent Banker - November 2024
